Academy of Management Journal
2005, Vol. 48, No. 4, 607 618.
A RELATIONAL PERSPECTIVE ON TURNOVER: EXAMINING
STRUCTURAL, ATTITUDINAL, AND BEHAVIORAL
PREDICTORS
KEVIN W. MOSSHOLDER
Louisiana State University
RANDALL P. SETTOON
Southeastern Louisiana University
STEPHANIE C. HENAGAN
Louisiana State University
This study examined whether structural, attitudinal, and behavioral variables of a
relational nature were predictive of employee turnover. Participants were a sample of
176 health care employees from a large public medical center. Using survival analysis
over a five-year time frame, we found that two such variables, network centrality and
interpersonal citizenship behavior, predicted turnover. Implications of taking a rela-
tional perspective toward turnover and other withdrawal behaviors found in organi-
zations are discussed.
Much of the research examining turnover has lens. Thus, the purpose of the present study was to
grown from March and Simon s (1958) notions con- address relevant relational variables having theo-
cerning the perceived desirability and ease of leav- retical links with turnover and test their predictive
ing one s job. Reviews of this research stream (Grif- potential.
feth, Hom, & Gaertner, 2000; Maertz & Campion,
1998) confirm that perceived desirability, often
measured as job satisfaction or other job-related
THEORETICAL RATIONALE
attitudes, is negatively associated with turnover.
Support for a relational perspective on organiza-
Ease of movement variables, principally assessed
tional withdrawal processes is rooted partly in the
through job alternative perceptions and job search
concept of social capital, which refers to the sum of
behavior, also predict turnover. Despite the wide
actual and potential resources available through
acceptance of this research lineage, recognition that
relationships that individuals have established
other factors besides work attitudes and job alter-
with others (Leana & Van Buren, 1999). In review-
natives may be important for understanding turn-
ing the value of social relationships for competitive
over has stimulated some to expand turnover re-
advantages, Uhl-Bien and her colleagues (2000)
search in new directions (e.g., Mitchell & Lee,
noted that low-quality relationships can have large
2001).
costs for organizations, among them higher turn-
The present study springs from what could be
over. Others have suggested that, as constituted in
viewed as a paradox in the turnover literature. Al-
relational networks, social capital may reduce turn-
though conceptual frameworks have cast turnover
over (Krackhardt & Hanson, 1993). For the most
within the context of relational processes (e.g., at-
part, however, such discussions have been framed
tachment, separation, exchange), most studies have
in terms of strategic organizational concerns that
not explicitly emphasized the effects of relational
only indirectly address individual processes un-
variables on turnover. The theoretical relevance of
derlying this mitigating influence.
a relational perspective on turnover has begun to
To understand at a more formative level how
receive greater attention (Dess & Shaw, 2001; Uhl-
relationships may affect turnover, it is instructive
Bien, Graen, & Scandura, 2000), establishing an
to consider underlying processes from which rela-
impetus to investigate turnover through a relational
tional ties emerge. Social exchange theory (Blau,
1964) implies that processes fostering workplace
ties influence individuals interacting with other
The helpful comments of David Harrison, Rodger Grif-
organization members. Social exchanges rest on the
feth, Dan Newman, and Robert Steel on an initial draft
manuscript are gratefully acknowledged. notion that gestures of goodwill will be recipro-
607
608 Academy of Management Journal August
cated at some future time. Although instrumental rea- to exhibit these characteristics would normally be
sons may also be involved, the benefits exchanged described as coworkers. Researchers have noted
often take on value because they symbolize the qual- that feelings of mutual obligation and expectations
ity of the relationships. Leaving such exchange rela- of reciprocity and supportiveness are hallmarks of
tionships may entail a psychic loss, making with- high-quality exchanges among peers and coworkers
drawal personally costly to individuals. (e.g., Sherony & Green, 2002). In keeping with no-
Maertz and Griffeth (2004) identifed constituent tions of interpersonal exchange, we refined our
forces (attachments to others in an organization) as search to emphasize variables connoting such rela-
one of eight distinctive motivational forces under- tional qualities in a coworker milieu.
lying  voluntary turnover (departure from an or- Our review found sufficient theoretical and em-
ganization despite having an opportunity to re- pirical support for associating the following rela-
main). These researchers note that even extensive tional variables with turnover: network centrality,
turnover models have overlooked the potential role perceived coworker support, felt obligation toward
of relationships with coworkers and leaders, which coworkers, and interpersonal citizenship behavior.
may partly explain why few turnover studies have Given the health care context in which the current
focused explicitly on relational constructs. We be- study was conducted, we expected these variables
lieve that the chances of organizational attachment to be especially salient. Because providing care in
grow concomitantly as individuals develop more such a work environment requires giving of oneself
extensive, higher-quality social networks. Kahn emotionally, replenishment in the form of rela-
(1998) suggested that even though people develop tional and emotional support from coworkers and
work-related connections to move closer to organ- other employees is important (Parker, 2002). Stud-
izational goals, connections are also formed for ies have suggested that the absence of such support
more complicated reasons. His relational systems can result in withdrawal from an organization (see,
perspective suggests that noncognitive elements, e.g., Kahn, 1993). Network centrality should be piv-
such as respect, warmth, and personal regard otal in such situations, as stronger ties with co-
should be acknowledged as essential in such rela- workers mean more opportunities for employees to
tionships. Characteristics of high-quality relational exchange  social wares. Perceived coworker sup-
systems enmesh individuals within a relational port taps the amount of peer assistance employees
web, making them less susceptible to forces that feel they have received, and felt obligation indi-
could dislodge them from their organization. Sev- cates the degree to which they feel obliged to recip-
eral researchers have described this circumstance rocate in kind. Finally, interpersonal citizenship
as embedding and have argued that it protects behavior indicates whether employees have actu-
against shocks that lead people to consider with- ally engaged in helping coworkers. Below, we ex-
drawal (e.g., Mitchell & Lee, 2001) and that gradu- plain further why each of these variables should be
ally erode positive organizational attachment (Burt, expected to affect employee turnover and provide
2001). As Maertz and Campion (1998) noted, posi- specific hypotheses regarding such effects. It
tive work relationships can influence individuals should be noted that we considered other variables
to remain with an organization despite their dislik- that did refer to coworkers but did not include
ing various features of it. them in the study, as they were more attitudinal
We reviewed the turnover literature through a than relational (e.g., satisfaction with coworkers).
relational lens, focusing on studies in which social Other potential predictors had relational overtones
exchange processes (Blau, 1964) or relational sys- deriving from group dynamics (e.g., group cohe-
tems elements (Kahn, 1998) were evident. Vari- sion). Because we were interested in the dynamics
ables connoting key relational processes such as of interpersonal relations rather than more diffuse
reciprocity, support, and mutual obligation were group processes, such variables were also excluded.
considered as potential predictors of turnover, as
were variables suggesting connectedness with oth-
Hypotheses
ers. Another important stricture in our literature
search concerned the locus of exchange. Opportu- Network centrality. Interpersonal networks de-
nities to form ties occur more readily in contexts lineate interaction patterns among individuals in
that are free of constraints imposed by status or organizations. These relationship structures facili-
power differences. Relational ties would also be tate shared meaning about goals and values and act
expected to arise more often among individuals as important identity referents for individuals
working proximally and among those who are in- learning what it means to be part of an organization
terdependent rather than independent in their (Bolino, Turnley, & Bloodgood, 2002). Broadly
tasks and goals. In the workplace, those most likely speaking, network centrality refers to the relative
2005 Mossholder, Settoon, and Henagan 609
numbers of direct and indirect links an individual support regarding social embedding effects. They
has with others comprising a social network. Al- found that relational job learning, which measures
though ways of measuring network centrality have the degree to which individuals learn about their
differed depending on the theoretical focus, all per- connectedness to other employees, was correlated
tain to interpersonal ties, making network central- negatively with employees  turnover intentions
ity relational by definition (Brass, 1995). Individu- (intentions to leave). In sum, theoretical and empir-
als with high centrality are more connected with ical research suggests that embeddedness and
others in their organizations, implying greater in- strong relational ties, as reflected by high network
volvement in assistance exchanges with coworkers, centrality, should inhibit attachment decay and re-
whether the purpose is work-related or more per- duce withdrawal tendencies. Therefore, we
sonal. Central individuals may also become valued hypothesize:
as sources of future assistance (Sparrowe, Liden,
Hypothesis 1. Network centrality is negatively
Wayne, & Kraimer, 2001), underscoring their long-
related to turnover.
er-term importance to the organizations and to the
members with whom they interact. Perceived coworker support. In work contexts,
Some previous studies have examined social net- social support refers to the care and consideration
works and organizational turnover, but more in that individuals receive from other organization
terms of structural rather than relational dynamics members. In the present study, coworker support
(e.g., Krackhardt & Porter, 1986). Recently, re- was considered as originating from interactions
searchers have begun to underscore how employee with others at the same organizational level. Facets
ties may affect their attachment to an organization. of social support have long been considered as
For example, Kahn (1998) argued that relationships potentially reducing individuals withdrawal in-
at work reflect not only necessary task-related tentions (e.g., Price & Mueller, 1986). However,
links, but a system of deeper attachments that serve only two studies investigating direct associations
collective emotional needs. To the degree that in- between coworker support and turnover could be
dividuals lack a supportive system of relationships, found. Iverson and Pullman (2000) determined that
they may become emotionally disengaged, with- coworker support predicted whether employees
draw psychologically then physically from cowork- would be laid off, but not whether they would
ers, and eventually leave the organization itself. voluntarily leave their organization. Iverson (1999)
A recent study by Burt (2001) suggested connec- suggested that coworker support would reduce
tions between individuals and an organization are turnover because employees could rely on cowork-
influenced by the degree of embedding they expe- ers as work became difficult, but he found no evi-
rience. Individuals having a higher number of ties dence of this effect.
to others affiliated with an organization are more These results are somewhat surprising given that
embedded and deem the organization to be of organizational support theory research (Rhoades &
greater importance to them. Burt (2001) found that Eisenberger, 2002) has widely indicated that per-
the degree of embedding inhibited the decay of ceived support from organizations and their agents
individuals organizational attachment. His results is negatively related with withdrawal behavior.
were consistent with an earlier study by McPher- The theoretical rationale for this finding is that
son, Popielarz, and Drobnic (1992) that showed demonstrations of care and consideration engender
more contacts in a social network led to longer emotional bonds between individuals and the
membership duration; more ties meant less sources of support. This bonding creates feelings of
turnover. social integration and identification that increase
Mitchell, Holtom, Lee, Sablynski, and Erez s individuals attachment to the support sources.
(2001) study examining a newly proposed organi- Logically, the association between coworker sup-
zational attachment construct, job embeddedness, port and turnover should parallel that found with
partially supports Burt s (2001) and McPherson other types of support. Work-based social support
and colleagues (1992) findings. Job embeddedness is usually extended by proximal individuals, and
predicted voluntary turnover in two organizations. thus exchanges of support should reverberate
 Links-to-the-organization, one of six components among coworkers, strengthening interpersonal
comprising job embeddedness, gauged personal bonds and attenuating shocks that encourage with-
connections among employees. This component it- drawal behavior. In sum, employees who are the
self correlated with turnover in one of the two object of coworkers caring and concern increase
organizations, partly corroborating the connection their direct attachment to such coworkers, and
between turnover and relational ties. Lankau and through them, indirect attachment to their organi-
Scandura s (2002) study yielded additional indirect zation. Thus, we hypothesize:
610 Academy of Management Journal August
Hypothesis 2. Perceived coworker support is both parties involved in help exchanges, reinforc-
negatively related to turnover. ing perceptions of mutual obligation. When it is
carried out to display the quality of ties with co-
Felt obligation toward coworkers. Social ex-
workers, interpersonal citizenship behavior may be
change theory (Blau, 1964) suggests individuals re-
largely noninstrumental because it is not intended
ceiving social wares develop feelings of obligation
to assist with specific tasks.
toward those supplying such wares and seek to
Individuals may also engage in interpersonal cit-
discharge those obligations through increased alle-
izenship behavior for instrumental reasons, such as
giance and extra performance efforts. Viewed as a
to help coworkers complete tasks or gain organiza-
prescriptive belief that one should care about an-
tional status. Such actions could influence individ-
other s well-being (Rhoades & Eisenberger, 2002),
uals leaving their organizations in the future. If
felt obligation underlies the give-and-take among
individuals view their interpersonal citizenship be-
coworkers in exchange relationships. Sherony and
havior as an investment that increases their value
Green (2002) offered that it is an important ingre-
to an organization and profession, they will be
dient in high-quality coworker relations. Felt obli-
more likely to stay and reap potentially ensuing
gation propels social relations because benefits ex-
rewards (Chen, Hui, & Sego, 1998). Thus, when
tended owing to obligations create feelings of
interpreted as an integrative work process, helping
investment and encourage individuals to work at
becomes a means of displaying behaviors that
preserving this incipient social capital. Because
could have desirable outcomes. For example, in an
bonds built between relational partners are rela-
engineering firm where helping was considered as
tively durable, feelings of obligation may continue
a communal good and a learning opportunity, in-
to influence relational ties after initial social wares
dividuals perceived that it benefited other employ-
have been requited.
ees, the organization, and their own careers (Perlow
Potential relations between felt obligation and
& Weeks, 2002). As with other relational variables
turnover have only recently begun to be examined.
examined in this study, interpersonal citizenship
As part of a study investigating reciprocation of
behavior and related constructs have not been fre-
perceived organizational support, Eisenberger,
quently considered in connection with turnover. In
Armeli, Rexwinkel, Lynch, and Rhoades (2001) as-
their study, Chen and colleagues (1998) found that
sessed relations between felt obligation toward an
altruism, an interpersonal helping facet of organi-
organization and employees withdrawal behavior.
zational citizenship behavior, correlated negatively
Felt obligation exhibited a significant, negative cor-
with turnover measured ten months later. Else-
relation with withdrawal behavior but did not exert
where, Aryee and Chay (2001) showed that citizen-
a significant effect in the context of their multivar-
ship behavior oriented toward helping individuals
iate model. Although they may not exert a strong
negatively correlated with turnover intentions in a
effect at a given juncture, felt obligations toward
union context. We thus hypothesize:
coworkers could elicit longer-term reciprocity in
Hypothesis 4. Interpersonal citizenship behav-
interpersonal exchanges, which would encourage
ior is negatively related to turnover.
closer relationships and greater interpersonal em-
beddedness. Given these expected relational ef-
fects, we hypothesize:
METHODS
Hypothesis 3. Felt obligation toward coworkers
Sample Characteristics
is negatively related to turnover.
Some of the data used in the present study were
Interpersonal citizenship behavior. Interper- collected as part of a larger study (Settoon & Moss-
sonal citizenship behavior refers to citizenship be- holder, 2002) focusing on precursors of interper-
havior directed toward coworkers and immediate sonal citizenship behavior; coworker support and
others and focuses on cooperative assistance for network centrality were examined in this capacity.
individuals in need (Settoon & Mossholder, 2002). Nearly five years after the initial data collection, we
From a relational perspective, behavior performed obtained permission to revisit a data collection site
to benefit coworkers indicates the depth of feeling and gather information on turnover in order to ex-
for and connection with others in an organization. amine whether salient relational variables pre-
This is especially so because interpersonal citizen- dicted turnover.
ship behavior occurs at individuals discretion; Study participants were employed at a regional
those who go out of their way to help coworkers are medical center in the southern United States. The
actively demonstrating an allegiance to them. Help- service-oriented work they performed required
ing behavior may evoke positive emotions from much communication and coordination among em-
2005 Mossholder, Settoon, and Henagan 611
ployees, providing a context where relational de- Adopting a relational cohesion approach (Brass,
pendencies and accompanying attitudes and be- 1995), we constructed a network centrality index as
haviors could surface. We used an employee survey per Ibarra (1993) by having respondents list up to
to collect information on key study variables, so- five employees (1) with whom they discussed what
ciometric relations, and demographics. A separate was going on in the organization and (2) whom they
survey was used to gather supervisory evaluations approached concerning work-related problems or
of employees interpersonal citizenship behavior. decisions. From these responses, we calculated  in-
Employees were asked to provide the last five digits degree centrality within the organization s com-
of their employee identification numbers so that munication and advice networks, respectively: the
their surveys could be matched with supervisors more coworkers choosing a focal employee, the
surveys. Participants were informed that their re- greater was that employee s in-degree centrality.
sponses were known only by the researchers and Assessing centrality for these two networks al-
would be kept confidential. Both supervisory and lowed consideration of interconnections having so-
nonsupervisory employees received surveys along cial (communication) as well as more work-related
with postage-paid return envelopes, which they (advice) connotations. As research has shown that
were to put in a collection bin in the administrative measures of centrality across networks are highly
offices or mail directly to us. correlated (r .78 in the present study), we com-
A total of 374 surveys were distributed, and 253 bined the in-degree centrality indexes for the two
were returned. Of these, 38 surveys could not be networks to form an overall measure.
used owing to either missing employee identifica- We developed a measure of perceived coworker
tion numbers or lack of matching supervisor sur- support ( .82) based on Eisenberger, Hunting-
veys; 215 surveys were usable, giving us a 58 per- ton, Hutchison, and Sowa s (1986) perceived organ-
cent response rate. There were no differences izational support scale. From their measure, we
between the usable and unusable survey groups on selected the six  highest-loading items gauging
tenure, gender, race, or education, but the average support and reworded them so that the referent was
age in the unusable group was nearly five years the responding employee s coworkers. Developed
lower (p .05). Given that tenure is the most mean- for the present study, the felt obligation toward
ingful demographic predictor of turnover (Griffeth coworkers measure had three items that reflected
et al., 2000), along with the absence of other demo- psychological feelings of indebtedness toward co-
graphic differences, the potential for response bias workers. Because these items were written for the
was judged as minimal. Human resources person- present study, we also assessed and found a coef-
nel helped to determine those employees whose ficient alpha of .74 for them using an independent
departures from the firm were involuntary (e.g., sample of 58 employees in an auxiliary services
dismissals), and these cases were eliminated from organization. Coefficient alpha in the hospital sam-
consideration. This constraint and missing data on ple was .67, despite an average item intercorrela-
other study variables reduced the sample to 176 tion of .40. As this circumstance may reflect scale
individuals. Of these 176 respondents, 78 percent length rather than a lack of internal consistency
were women, 72 percent were Caucasian, and 27 (see Cortina, 1993), we decided to retain this mea-
percent were African American. Average respon- sure. Finally, we used a previously developed mea-
dent age was 37.43 years, and average organization- sure (Settoon & Mossholder, 2002) to assess inter-
al tenure was 5.04 years. personal citizenship behavior ( .96). Eight of 14
items assessed passive support provided to others,
and 6 measured active assistance to those in need.
Measures
Although interpersonal citizenship behavior has
Predictor variables. Three survey formats were been examined previously as comprising two fac-
used: (1) Likert-type scales ranging from 1, ets, the thrust of the present study was the predic-
 strongly disagree, to 5,  strongly agree, (2) so- tive power of interpersonal citizenship behavior
ciometric questions, and (3) demographic ques- overall. Therefore, we collapsed all of the items to
tions. Employees self-reported data on all variables form one scale.
with the exception of interpersonal citizenship be- Criterion variable. The effects of relational vari-
havior, for which supervisors provided data, and ables on turnover may surface more slowly than
network centrality, which was measured on the those of other types of variables (e.g., economic
basis of coworker responses. As appropriate, we conditions) because relationships are less formally
averaged responses to items to create the scales. determined aspects of work and require time to
The texts of all scale items and sociometric ques- develop. At least two studies have employed a five-
tions appear in the Appendix. year survival window in examining socially ori-
612 Academy of Management Journal August
ented variables impacts on turnover (e.g., Iverson survival analysis approach, Cox regression analy-
& Pullman, 2000; O Reilly, Caldwell, & Barnett, sis. The proportional hazards assumption of Cox
1989). Also, Sheridan (1985) noted that most vol- regression requires that the effect of a predictor
untary turnover on nursing staffs occurs within 60 variable does not change over time. In other words,
months. Nearly five years after the initial survey, in no interaction between time and the predictor
keeping with Griffeth et al. s (2000) recommenda- should be observed. One means of testing this as-
tion to use a time frame appropriate for the turn- sumption is to add time-by-predictor interaction
over context, human resources personnel were terms to an equation already containing predictor
asked to provide turnover data on the 215 individ-  main effect terms (Harrison, 2002). We created
uals for whom identification information and inter- the necessary interaction terms and tested them for
personal citizenship behavior ratings were avail- significance following standard procedures (see
able. Their employment status was verified, and Tabachnick & Fidell, 2001). None of the interaction
separation dates for those no longer employed pro- terms were significant, and thus the assumption
vided. We measured the survival window in was met. Moreover, for each predictor, we plotted
months. We considered voluntary turnover to have Schoenfeld residuals against time. These residuals
occurred when employees left despite having an were distributed fairly evenly in a band around
opportunity to continue with the organization zero for all predictors, providing visual confirma-
(Maertz & Campion, 1998), and the codings were 0 tion of the proportional hazards assumption (Hess,
for  stayed and 1 for  left. 1995).
Control variables. Organizational tenure and
age, measured in years, were controlled as both
have been found to correlate with turnover (Grif-
RESULTS
feth et al., 2000). Additionally, researchers exam-
ining health care samples (e.g., Iverson & Pullman, A confirmatory factor analysis involving the per-
2000) have suggested that gender be controlled as ceptual (perceived coworker support, felt obliga-
women are overrepresented in such contexts. Gen- tion toward coworkers), attitudinal (job satisfac-
der was coded  male, 0, and  female, 1. Finally, tion), and behavioral (interpersonal citizenship
we used a four-item general job satisfaction mea- behavior) measures was conducted as a means of
sure ( .79) as a control because job satisfaction assessing their viability as separate constructs. To
has been consistently associated with turnover increase indicator stability (West, Finch, & Curran,
(Griffeth et al., 2000). This measure comprised two 1995) and meet sample size guidelines for param-
items from the Michigan Organizational Assess- eter estimation, we employed the single-factor
ment Questionnaire (Cammann, Fichman, Jenkins, method (Landis, Beal, & Tesluk, 2000) to form a
& Klesh, 1983) and two items from Brayfield and reduced set of indicators for each latent variable
Rothe s (1951) index. Controlling job satisfaction (excepting felt obligation, which comprised only
helped assure that relations between turnover and three indicators). This procedure reduced the num-
correlates of satisfaction with work, such as age and ber of indicators per latent variable to three (for
tenure, were not confounded. coworker support and job satisfaction) or four (for
interpersonal citizenship behavior). The compara-
tive fit index (CFI), root-mean-square error of ap-
Analyses
proximation (RMSEA), and standardized root-
Because relationships are established through re- mean-square residual (SRMR) were used in judging
peated interactions with others, relational variables fit. Utilizing LISREL 8, we tested a model with four
may exert their influence gradually in connection latent variables. All indicators loaded significantly
with turnover rather than at a single point in time. on their respective latent variables. The resultant
Survival analysis techniques incorporate time- indexes suggested an acceptable level of fit (CFI
based dynamics befitting relational variables and .99, RMSEA . 04, and SRMR .05).
are well suited to modeling such processes. Sur- Table 1 presents descriptive statistics for study
vival analysis methods assess turnover in terms of variables, along with correlations among the vari-
the conditional probability of a member s leaving ables. As would be expected, the correlation be-
an organization; this probability varies with the tween age and tenure was moderately strong. Also,
duration of organizational membership. Several re- coworker support and felt obligation toward co-
searchers have noted the general advantages and workers were significantly correlated, corroborat-
increasing use of survival analysis on turnover data ing the idea that perceived support from coworkers
(e.g., Harrison, 2002). may engender feelings of social indebtedness. Most
We analyzed data using a proportional hazards of the remaining correlations were relatively low,
2005 Mossholder, Settoon, and Henagan 613
TABLE 1
Descriptive Statistics and Correlations among the Variablesa
Variable Mean s.d. 1 2 3 4 5 6 7 8
1. Organizational tenure 5.04 5.56
2. Gender 0.78 0.41 .06
3. Age 37.43 10.59 .43** .00
4. Job satisfaction 4.17 0.56 .04 .10 .12
5. Interpersonal citizenship behavior 3.84 0.74 .02 .16* .04 .07
6. Coworker support 3.69 0.61 .02 .04 .01 .24** .14
7. Felt obligation toward coworkers 2.73 0.95 .09 .01 .11 .06 .00 .28**
8. Network centrality 1.54 1.51 .15* .12 .01 .06 .10 .01 .18*
9. Turnoverb 0.36 0.48 .15* .02 .14 .13 .13 .04 .08 .22**
a
n 176.
b
Indicates the turnover rate for the sample after five years.

p .10
*p .05
**p .01
indicating a degree of empirical differentiation shows pseudo-R2 for the model containing controls
among the variables. only and the model containing all the variables.
Figure 1 shows plots of the estimated survival
and hazard functions. Both functions exhibit a con-
DISCUSSION
sistent linear trend throughout most of the study
time frame. Toward the final months of the study, The turnover literature has been dominated by
slight plateauing occurs for both functions. Table 2 research on how work attitudes, especially job sat-
reports results of the Cox regression analyses. We isfaction, progress into turnover. Although this re-
used likelihood ratio tests to examine the effects of search stream has contributed much to the litera-
entering control and predictor variable blocks into ture, there has been an increased interest in the role
the model, and Wald statistics to determine that less traditional variables (e.g., nonwork) and
whether individual variables were statistically sig- precipitating shocks play in withdrawal processes
nificant. In the first step, only the control variables (Mitchell & Lee, 2001). The present study repre-
were entered into the regression equation. A likeli- sents an initial attempt to determine whether rela-
hood ratio test indicated there was not a significant tional variables specifically, network centrality,
(p .05) change in model fit. Addition of the perceived coworker support, felt obligation toward
relational variables as a set in the second step of the coworkers, and interpersonal citizenship behav-
analysis resulted in a significant chi-square change ior can contribute in this regard.
in fit ( 2 15.97, df 4, p .003). Both network centrality and interpersonal citi-
Examining the regression coefficients with all the zenship behavior were significantly related to turn-
variables entered into the model, one can see that a over over the five-year study window, and their
negative association between job satisfaction and effects went above and beyond the effects of job
turnover was marginally significant (p .06). As satisfaction. Regarding network centrality, it ap-
for the relational variables in the full model, Table pears that how individuals are interconnected af-
2 shows that both network centrality (p .01) and fects their leaving an organization. Those forming a
interpersonal citizenship behavior (p .05) were greater number of ties with coworkers become more
significantly associated with turnover. Coworker embedded and more likely to identify with those
support and felt obligation toward coworkers did around them (Burt, 2001). Such social grounding
not predict turnover. We computed pseudo-R2s dampens the effect of real or perceived shocks that
(Harrison, 2002) to show the relative improvement may give rise to turnover. Close identification with
in the association between survival and each sig- others is often associated with perceived similarity
nificant predictor, adjusting for all other predictors. and more frequent communication, which in turn
Adding network centrality to a model containing may reduce the likelihood of turnover (Brass,
all other study variables increased this statistic by 1995). Interestingly, felt obligation toward cowork-
.04, whereas adding interpersonal citizenship be- ers was negatively correlated with network central-
havior to a model containing all other study vari- ity. Providing others with advice and communica-
ables increased pseudo-R2 by .03. Table 2 also tion may lead highly central individuals to feel
614 Academy of Management Journal August
FIGURE 1
Estimated Survival and Hazard Functions
they have discharged their personal obligations to timating that relational exchanges may benefit both
coworkers, and it may indirectly indicate their partners. Thus, interpersonal citizenship behavior
awareness of the social investments they have can be viewed as a social workplace investment
made. that improves the lot of those receiving help while
The association between interpersonal citizen- increasing the chances that help giving will be re-
ship behavior and turnover is intriguing because ciprocated in the future. Alternatively, individuals
interpersonal citizenship behavior is directed at who have doubts about remaining with an organi-
coworkers rather than an organization. How indi- zation or who are actively contemplating leaving
viduals interact with coworkers in this case, help- may be less likely to help coworkers, knowing that
ing them may presage turnover tendencies and their own exits will nullify future return benefits.
represent a relational manifestation of their organ- The positive correlation between coworker sup-
izational attachment. Relationship quality has been port and felt obligation toward coworkers suggests
found to be positively associated with both helping that receiving social wares may stimulate a need to
and help seeking (Anderson & Williams, 1996), in- reciprocate. However, the hypothesized relations of
2005 Mossholder, Settoon, and Henagan 615
TABLE 2
Results of Cox Regression Analysesa
Step 1 Step 2
Predictors b s.e. eb b s.e. eb
Organizational tenure 0.05 0.03 0.95 0.04 0.04 0.96
Gender 0.00 0.32 1.00 0.03 0.33 0.97
Age 0.01 0.01 0.99 0.02 0.01 0.98
Job satisfaction 0.30 0.20 0.74 0.43 0.23 0.65
Controls-only pseudo-R2 .04
Network centrality 0.29** 0.11 0.75
Coworker support 0.31 0.24 1.36
Felt obligation toward coworkers 0.02 0.14 1.02
Interpersonal citizenship behavior 0.40* 0.17 0.68
Full-model pseudo-R2 .13
a
Unstandardized regression coefficients, standard errors, and exponentiated regression coefficients are shown.

p .10
* p .05
** p .01
these two perceptual predictors with turnover did [2004]) may share explanatory variance with tradi-
not emerge. One reason for this may be that the tional attitudinal variables (i.e., job satisfaction, or-
five-year period between initial measurement and ganizational commitment) in predicting turnover.
turnover in the current study is longer than the Others have noted that job satisfaction is a key
in-between period of typical turnover studies (see variable in predicting turnover (Mueller, Boyer,
Griffeth et al., 2000), and perceptual variables may Price, & Iverson, 1994). Having assessed job satis-
not be suited to such lengthy time frames. Also, faction in our study, we were able to control for its
researchers have noted that even when predictive, effects. Unfortunately, because of constraints on
perceptual variables tend to exhibit modest effect survey length, information on other traditional
sizes in connection with turnover. turnover predictors could not be included. Such
This study provides direct support for the idea variables should be incorporated in future studies
that relationships matter, and it corroborates con- investigating relational predictors of turnover.
tentions (e.g., Bolino et al., 2002; Uhl-Bien et al., It may also be instructive to examine explicitly
2000) that their development influences important relational variables in conjunction with more dif-
organizational outcomes. Of course, it should be fuse group-level processes that have been investi-
noted that the relational variables examined in our gated in past research. For example, group cohe-
study represent only one of eight motive categories sion has been found to correlate negatively with
that drive employees decisions about whether to withdrawal phenomena (see Griffeth et al., 2000;
remain organization members. Future research in- O Reilly et al., 1989; Sheridan, 1985), and it is a
volving relational variables should include vari- global reflection of the degree of individuals attrac-
ables representing other motive categories, espe- tion to a group. As noted above, although group-
cially ones that have been investigated in previous level processes are not considered to be explicitly
studies (e.g., affective, alternative, or behavioral relational, they have a theoretical kinship with re-
forces; see Maertz and Griffeth [2004]). Including lational variables. Including such variables in fu-
other categories would better gauge the predictive ture studies would represent a logical extension of
strength of relational variables against that of more the present study.
traditional turnover predictors, such as organiza- Some caveats should be mentioned regarding our
tional attitudes (e.g., affective organizational com- findings. First, the data were collected in a health
mitment), external environmental factors (e.g., per- care context, a type of work setting that is in some
ceived work alternatives), and behavioral correlates respects more oriented toward relationships than
(e.g., absences). A recent study suggests that some are other work settings. Replication in a variety of
relational constructs (e.g., on-the-job-embedded- other organizations is needed, as relational vari-
ness; Lee, Mitchell, Sablynski, Burton, and Holtom ables may operate differently in other types of con-
616 Academy of Management Journal August
Aryee, S., & Chay, Y. W. 2001. Workplace justice, citi-
texts. For example, some researchers have found
zenship behavior, and turnover intentions in a union
that in industries where turnover is very high (fast-
context: Examining the mediating role of perceived
food restaurants), strong ties may exacerbate turnover
union support and union instrumentality. Journal of
among employees with similar roles (see Krackhardt
Applied Psychology, 86: 154  160.
& Porter, 1986). Second, our study was retrodictive,
Blau, P. 1964. Exchange and power in social life. New
in that it did not begin with a cohort of new employ-
York: Wiley.
ees whose survival rates were tracked going forward.
As such, it is not a survival study in the purest sense.
Bolino, M. C., Turnley, W. H., & Bloodgood, J. M. 2002.
Third, some of the measures were based on existing
Citizenship behavior and the creation of social cap-
scales as opposed to being developed specifically for ital in organizations. Academy of Management Re-
view, 27: 505 522.
the study. Closer consideration of measures focusing
on relational constructs is needed in the future.
Brass, D. J. 1995. A social network perspective on human
We suggest that relational predictors may contain
resources management. In G. R. Ferris (Ed.), Re-
unique information about withdrawal and hold
search in personnel and human resources man-
promise for solving another piece of the turnover agement, vol. 13: 39  79. Greenwich, CT: JAI Press.
puzzle. Our results suggest that when researchers
Brayfield, A., & Rothe, H. 1951. An index of job satisfac-
study withdrawal phenomena, interpersonal rela-
tion. Journal of Applied Psychology, 35: 307 311.
tionships may be critical to consider along with the
Burt, R. S. 2001. Attachment, decay, and social network.
more typically emphasized person characteristics.
Journal of Organizational Behavior, 22: 619  643.
Managing turnover through relational variables
Cammann, C., Fichman, M., Jenkins, G., & Klesh, J. 1983.
will require new and innovative approaches.
Assessing the attitudes and perceptions of organiza-
Rather than simply focusing on individuals to
tional members. In S. Seashore, E. Lawler, P. Mirvis,
thwart turnover, organizations may benefit from
& C. Cammann (Eds.), Assessing organizational
considering broader circumstances, such as how
change: 71 138. New York: Wiley.
embeddedness may protect individuals from
Chen, X., Hui, C., & Sego, D. J. 1998. The role of organi-
events that function as turnover-inducing shocks.
zational citizenship behavior in turnover: Conceptu-
Taking steps to fortify the broader relational cli-
alization and preliminary tests of key hypotheses.
mate could involve programmatic actions aimed at
Journal of Applied Psychology, 83: 922 931.
preventing both physical and emotional disengage-
Cortina, J. M. 1993. What is coefficient alpha? An exam-
ment at work (Kahn, 1998). Such strengthening
ination of theory and applications. Journal of Ap-
may not only retard turnover, but may also develop
plied Psychology, 78: 98  104.
relational social capital that could improve organ-
Dess, G. G., & Shaw, J. D. 2001. Voluntary turnover, social
ization-wide productivity (Dess & Shaw, 2001).
capital, and organizational performance. Academy
Systematic efforts should be directed at developing
of Management Review, 26: 446 456.
links among individuals, an idea congruent with the
growing emphasis on personal learning in organiza- Eisenberger, R., Armeli, S., Rexwinkel, B., Lynch, P., &
Rhoades, L. 2001. Reciprocation of perceived organ-
tions (cf. Lankau & Scandura, 2002). A key facet of
izational support. Journal of Applied Psychology,
what individuals learn as organization members may
86: 42 51.
be how they are interconnected and can work with
others to develop their own skills and careers.
Eisenberger, R., Huntington, R., Hutchison, S., & Sowa,
For some time, researchers have proposed that D. 1986. Perceived organizational support. Journal
of Applied Psychology, 71: 500 507.
 the people make the place (e.g., Schneider, 1987).
We further suggest that relational ties people form
Griffeth, R. W., Hom, P. W., & Gaertner, S. 2000. A meta
while working together may be the ties that bind.
analysis of antecedents and correlates of employee
Greater attention to relational systems and behav-
turnover: Update, moderator tests, and research im-
iors that strengthen interpersonal ties may inform plications for the next millennium. Journal of Man-
agement, 26: 463 488.
organizations efforts to reduce undesired turnover
and improve overall effectiveness. Hopefully, the
Harrison, D. A. 2002. Structure and timing in limited
present study will encourage future research efforts
range dependent variables: Regression models for
having these goals.
predicting if and when. In F. Drasgow & N. Schmitt
(Eds.), Measuring and analyzing behavior in organ-
izations: Advances in measurement and data anal-
REFERENCES
ysis: 446 497. San Francisco: Jossey Bass.
Anderson, S. E., & Williams, L. J. 1996. Interpersonal, job, Hess, K. R. 1995. Graphical methods for assessing viola-
and individual factors related to helping processes at tions of the proportional hazards assumption in Cox
work. Journal of Applied Psychology, 81: 282 296. regression. Statistics in Medicine, 14: 1707 1723.
2005 Mossholder, Settoon, and Henagan 617
Ibarra, H. 1993. Network centrality, power and innova- Mitchell, T. R., Holtom, B., Lee, T. W., Sablynski, C., &
tion involvement: Determinants of technical and ad- Erez, M. 2001. Why people stay: Using job embed-
ministrative roles. Academy of Management Jour- dedness to predict voluntary turnover. Academy of
nal, 36: 471 501.
Management Journal 44: 1102 1121.
Iverson, R. D. 1999. An event history analysis of em- Mitchell, T. R., & Lee, T. W. 2001. The unfolding model
ployee turnover: The case of hospital employees in
of voluntary employee turnover and job embedded-
Australia. Human Resource Management Review,
ness: Foundations for a comprehensive theory of
9: 397 418.
attachment. In B. M. Staw & R. I. Sutton (Eds.),
Research in organizational behavior, vol. 23: 189
Iverson, R. D., & Pullman, J. A. 2000. Determinants of
246. Greenwich, CT: JAI Press.
voluntary turnover and layoffs in an environment of
repeated downsizing following a merger: An event
Mueller, C., Boyer, E., Price, J. L., & Iverson, R. D. 1994.
history analysis. Journal of Management, 26: 977
Employee attachment and noncoercive conditions of
1003.
work. Work and Occupations, 21: 179 212.
Kahn, W. A. 1993. Caring for the caregivers: Patterns of
O Reilly, C. A., Caldwell, D., & Barnett, W. 1989. Work
organizational caregiving. Administrative Science
group demography, social integration, and turnover.
Quarterly, 38: 539  563.
Administrative Science Quarterly, 34: 21 37.
Kahn, W. A. 1998. Relational systems at work. In L. L.
Parker, V. A. 2002. Connecting relational work and work-
Cummings & B. M. Staw (Eds.), Research in organ-
group context in caregiving organizations. Journal of
izational behavior, vol. 20: 39  76. Greenwich, CT:
Applied Behavioral Science, 38: 276  297.
JAI Press.
Perlow, L., & Weeks, J. 2002. Who s helping whom? Lay-
Krackhardt, D., & Hanson, J. R. 1993. Informal networks:
ers of culture and workplace behavior. Journal of
The company behind the chart. Harvard Business
Organizational Behavior, 23: 345 361.
Review, 71(4):104  111.
Price, J. L., & Mueller, C. 1986. Absenteeism and turn-
Krackhardt, D., & Porter, L. W. 1986. The snowball effect:
over of hospital employees. Greenwich, CT: JAI
Turnover embedded in communication networks.
Press.
Journal of Applied Psychology, 71: 50  55.
Rhoades, L., & Eisenberger, R. 2002. Perceived organiza-
Landis, R., Beal, D., & Tesluk, P. 2000. A comparison of
tional support: A review of the literature. Journal of
approaches to forming composite measures in struc-
Applied Psychology, 87: 698  714.
tural equation models. Organizational Research
Schneider, B. 1987. The people make the place. Person-
Methods, 3: 186  207.
nel Psychology, 40: 437 453.
Lankau, M. J., & Scandura, T. A. 2002. An investigation of
personal learning in mentoring relationships: Con- Settoon, R. P., & Mossholder, K. W. 2002. Relationship
quality and relationship context as antecedents of
tent, antecedents, and consequences. Academy of
Management Journal, 45: 779  790. person and task focused interpersonal citizenship
behavior. Journal of Applied Psychology, 87: 255
Leana, C. R., & Van Buren, H. J. 1999. Organizational
267.
social capital and employment practices. Academy
of Management Review, 24: 538 555. Sheridan, J. E. 1985. A catastrophe model of employee
withdrawal leading to low job performance, high
Lee, T. W., Mitchell, T. R., Sablynski, C., Burton, J., &
absenteeism, and job turnover during the first year of
Holtom, B. 2004. The effects of job embeddedness on
employment. Academy of Management Journal,
organizational citizenship, job performance, voli-
28: 88  109.
tional absences, and voluntary turnover. Academy
of Management Journal, 47: 711 722.
Sherony, K. M., & Green, S. G. 2002. Coworker exchange:
Relationships between coworkers, leader-member
Maertz, C. P., & Campion, M. A. 1998. 25 years of volun-
exchange, and work attitudes. Journal of Applied
tary turnover research: A review and critique. In
Psychology, 87: 542 548.
C. L. Cooper & I. T. Robertson (Eds.), International
review of industrial and organizational psychol-
Sparrowe, R. T., Liden, R. C., Wayne, S. J., & Kraimer,
ogy, vol. 13: 49  81. Chichester, England: Wiley.
M. L. 2001. Social networks and the performance of
individuals and groups. Academy of Management
Maertz, C. P., & Griffeth, R. W. 2004. Eight motivational
Journal, 44: 316  325.
forces and voluntary turnover: A theoretical synthe-
sis with implications. Journal of Management, 30:
Tabachnick, B. G., & Fidell, L. S. 2001. Using multivar-
667 683.
iate statistics. Boston: Allyn & Bacon.
March, J. G., & Simon, H. A. 1958. Organizations. New
Uhl-Bien, M., Graen, G. B., & Scandura, T. A. 2000.
York: Wiley.
Implications of leader-member exchange (LMX) for
McPherson, J. M., Popielarz, P. A., & Drobnic, S. 1992. strategic human resource management systems: Re-
Social networks and organizational dynamics. lationships as social capital for competitive advan-
American Sociological Review, 57: 153 170. tage. In G. R. Ferris (Ed.), Research in personnel
618 Academy of Management Journal August
and human resources management, vol. 18: 137 Helps coworkers with difficult assignments, even
185. Greenwich, CT: JAI Press. when assistance is not directly requested.
Assists coworkers with heavy workloads, even though
West, S. G., Finch, J. F., & Curran, P. J. 1995. Structural
it is not part of his/her job.
equation models with nonnormal variables: Prob-
Goes out of his/her way to help coworkers with work-
lems and remedies. In R. H. Hoyle (Ed.), Structural
related problems.
equation modeling: 56 75. London: Sage.
Takes on extra responsibilities in order to help co-
workers when things get demanding at work.
Helps coworkers who are running behind in their work
APPENDIX
activities.
Items for Study Measures
Network Centrality
Items and instructions are given verbatim.
The following instructions preceded five lines that
Felt Obligation toward Coworkers
each included a space in which a respondent could
enter a name, followed by the words  talk and  ad-
I often feel like I owe my coworkers.
vice in capital letters.
My coworkers have done things for me that I feel I
should repay them for.
In the five blank lines below, print the first and last
Sometimes I do favors for my coworkers because I feel
name of five employees that you interact with on a
I am obligated to.
regular basis. For each employee that you list, circle
TALK if you discuss what is going on in the organiza-
Perceived Coworker Support
tion with that person. Circle ADVICE if this person is
My coworkers really care about my well-being.
an important source of professional advice when you
My coworkers are willing to extend themselves in
have a problem or decision to make. There may be
order to help me perform my job the best I can.
instances where you may circle more than one choice
Even if I did the best job possible, my coworkers would
for a particular employee. There also may be instances
fail to notice. (reverse-scored)
where you circle none for a particular employee.
My coworkers care about my general satisfaction at
work.
My coworkers show very little concern for me.
(reverse-scored)
My coworkers care about my opinions.
Kevin W. Mossholder ([e-mail]) is the H. N.
Saurage, Jr./Community Coffee Chair and a professor in
Job Satisfaction
the Management Department at Louisiana State Univer-
All in all, I am satisfied with my job.
sity. He received his Ph.D. from the University of Ten-
In general, I like working here.
nessee Knoxville. His primary research interests center
I like my job better than the average worker does.
on interpersonal workplace interactions, the effects of
I am satisfied with my job for the time being.
these interactions on organizational outcomes, and con-
textual issues that shape such effects.
Interpersonal Citizenship Behaviors
Randall P. Settoon is the department chair and a profes-
This employee. . .
sor of management at Southeastern Louisiana University.
Takes time to listen to coworkers problems and wor-
He received his Ph.D. from Louisiana State University.
ries.
His primary research interests include organizational cit-
Takes a personal interest in coworkers.
izenship behavior, social exchange in the workplace, and
Always goes out of the way to make newer employees
trust.
feel welcome in the work group.
Shows genuine concern and courtesy toward cowork-
Stephanie C. Henagan is currently an assistant professor
ers, even under the most trying business or personal
in the Management Department at Northern Illinois Uni-
situations.
versity and is completing her Ph.D. in organizational
Compliments coworkers when they succeed at work.
behavior and human resource management at Louisiana
Tries to cheer up coworkers who are having a bad day.
State University. Her current research interests include
Makes an extra effort to understand the problems faced
interpersonal dynamics in the workplace, the effects of
by coworkers.
social comparison processes on achievement, and affir-
Listens to coworkers when they have to get something
mative action attitudes.
off their chest.
Helps coworkers with work when they have been ab-
sent.